Aria Everyday Activities Dataset

We present Aria Everyday Activities (AEA) Dataset, an egocentric multimodalopen dataset recorded using Project Aria glasses. AEA contains 143 dailyactivity sequences recorded by multiple wearers in five geographically diverseindoor locations. Each of the recording contains multimodal sensor datarecorded through the Project Aria glasses. In addition, AEA provides machineperception data including high frequency globally aligned 3D trajectories,scene point cloud, per-frame 3D eye gaze vector and time aligned speechtranscription. In this paper, we demonstrate a few exemplar researchapplications enabled by this dataset, including neural scene reconstruction andprompted segmentation. AEA is an open source dataset that can be downloadedfrom https://www.projectaria.com/datasets/aea/. We are also providingopen-source implementations and examples of how to use the dataset in ProjectAria Tools https://github.com/facebookresearch/projectaria_tools.